RSRudraksh Syal

Now & Lab

Learning in public, without pretending the work is finished.

A living view of the ideas I am studying, testing, and connecting to a decade of practical data-science and ML experience.

Current thesis

Intelligent systems become useful when their behavior is observable, evaluable, and bounded by the decision they serve.

Current explorations

Four threads I’m actively connecting.

01Active focus

Agentic AI

Moving beyond demos toward systems with explicit tools, bounded autonomy, observable state, and recoverable failure modes.

Goal→Tools→Check→Act
  • Tool-use reliability
  • Planning vs workflows
  • Human checkpoints
02Active focus

Evaluation

Designing evidence for multi-step AI behavior: task success, trajectory quality, groundedness, cost, latency, and operational risk.

Task→Trace→Score→Learn
  • Agent trajectories
  • Failure taxonomies
  • Production feedback
03Deepening

Deep Learning

Revisiting representation learning and modern architectures to strengthen the bridge between applied ML experience and emerging AI systems.

Input→Representation→Task
  • Transformers
  • Representation learning
  • Fine-tuning choices
04Continuing practice

Reliable ML Systems

Keeping the production fundamentals visible: monitoring, feedback loops, baseline comparisons, deployment constraints, and explainable guardrails.

Train→Deploy→Monitor→Improve
  • MLOps
  • Drift
  • Decision interfaces

Open questions

The lab begins with questions worth testing.

  1. 01

    How should multi-step agents be evaluated when the final answer hides a poor trajectory?

  2. 02

    Where should deterministic workflows end and agent autonomy begin?

  3. 03

    How can LLM interfaces make pricing and optimization systems easier to use without hiding their constraints?

  4. 04

    How can product design make complex technical reasoning easier to inspect and trust?

What this page is—and is not

A record of direction, not a claim of mastery.

Mature work belongs in the portfolio. Early experiments, reading notes, design sketches, and hypotheses belong here. As projects become substantive, they can graduate into full case studies.